index theory on curves k¿(x) -» kg(x). · 2018. 11. 16. · 1. introduction. in [4] baum and...
TRANSCRIPT
TRANSACTIONS OF THEAMERICAN MATHEMATICAL SOCIETYVolume 288. Number 2, April 1985
INDEX THEORY ON CURVES
by
peter haskell1
Abstract. This paper constructs from the 9-operator on the smooth part of a
complex projective algebraic curve a cycle in the analytically defined K homology of
the curve. The paper identifies the corresponding cycle in the topologically defined K
homology.
1. Introduction. In [4] Baum and Douglas generalize the Atiyah-Singer index
theorem. They give analytic and topological definitions of the K homology group
KU0(X) for any finite simplicial complex X. They call these groups K£(X) and
K¿(X), and using the Dirac operator they construct the natural isomorphism
K¿(X) -» Kg(X).
When X is a smooth closed manifold, examples of Baum and Douglas's work arise
from elliptic differential operators on X. When X is not a smooth manifold,
intrinsically defined elements of K£(X) are harder to find. Thus for X not a smooth
manifold two natural problems are:
(a) define elements of Kq( X) as intrinsically as possible;
(b) for each such element find the associated element in K'0( X).
This paper solves these problems when X is an arbitrary complex projective
algebraic curve, which for simplicity is assumed to be without isolated points. §2
describes the parts of Baum and Douglas's work that are needed here. §3 introduces
the necessary function spaces. §4 describes the Laplacian and the applications of the
functional calculus used later. §5 constructs a cycle for K£(X) from the 3-operator
on the smooth part of X. §6 constructs a corresponding cycle in K¿(X) from the
desingularization of X.
I thank Paul Baum and Jeff Cheeger for their invaluable help.
2. K homology. The details of the material in this section can be found in [4].
A cycle for Kj¡(X) is given by a 5-tuple (H0, \p0, Hx, \px, T) satisfying:
(a) H0 and Hx are separable Hilbert spaces;
(b) t/>,: C(X) -> Se(H¡) is a unital algebra *-homomorphism for i = 0,1 (C(X) =
continuous C-valued functions on X, and ¿¡?(H¡) = bounded linear operators on
Received by the editors April 16. 1984. The results in this paper were presented to the 797th meeting of
the American Mathematical Society on October 31, 1982.
1980 Mathematics Subject Classification. Primary 55N15, 58G10, 58G05; Secondary 46L05, 58G16.
Key words and phrases. K homology, index of elliptic operator, complex algebraic curve.
'This paper is based on the author's Ph. D. thesis. Brown University, 1982.
©1985 American Mathematical Society
0002-9947/85 $1.00 + $.25 per page
591
License or copyright restrictions may apply to redistribution; see https://www.ams.org/journal-terms-of-use
592 PETER HASKELL
(c) T: H0 -» Hx is a bounded Fredholm operator with T° 4>0(f) — ̂ Áf)0 T
compact for each/g C(X).
The equivalence relation imposed on these cycles to get K{¡(X) is in the same spirit.
These cycles push forward by the pullback of continuous functions.
A cycle for K¿(X) is a triple (M, E, f) satisfying:
(a) Af is a compact Spin' manifold without boundary;
(b) E is a complex vector bundle on M;
(c) / is a continuous map M -> X.
The equivalence relation, which includes bordism, leading to K¿(X) is in the same
spirit.
The cap product K°(X)X K0(X) -> K0(X) is defined for both definitions of the
K homology group. If ( M, E, f) is a cycle for K¿( X) and F is a vector bundle on X,
then [F] n [(M, E, /)] = [(M, E 9 f*F, /)]. Let (H0, ¡p0, Hx, t^, T) be a cycle for
K£( X), and let F be a vector bundle on X. There is an integer n and a continuous
map P: X —» {Projections on C"} such that F is isomorphic to the bundle of images
of the projections. Denote by P'y the entry functions for P with respect to the
standard basis {ex,...,en} of C". Define projections P0 on H0 ® C" and Pt on
Hx 9 C"by
p0(j®e,)= I, (MPij))(y) ® ej7 = 1
and
n
/>,(*?»*,.)= E(^i(^'7))(w)®ey.
Then
[F] n [(H0, *0, fflf *lf P)] = [(P0(//0 ® C"), P0(^o ® 1)F0, F^ ® C"),
P1(*19 1)P1,P1(T9 1)P0)]
(see [1] and [9]). Despite the numerous choices, these cap products are well defined
on the K homology and K cohomology groups.
The push forward to Appoint) = Z of [(H0, x¡/0, Hx, \px, T)] is index(F). The push
forward to tf0(point) of [(M, E, /]) is (ch(E) U Td(TM))[M].
The isomorphism Kq(X) -* K£(X) uses the Dirac operators of manifolds appear-
ing in the cycles for K¿(X). In particular if Af is a nonsingular complex projective
curve, 1 the trivial line bundle on M and \p0 and \pl pointwise multiplication by
functions, then the classes of (Af, 1, identity) and (L2(M), ip0, L2(M, T*), ipi, 9 °
(1 - AM)~1/2) are associated to each other by the isomorphism K'0(M) -* K£(M).
3. Function spaces on curves. The following notation will be used henceforth:
A' is a complex projective algebraic curve;
77 : X —> X is the desingularization of X;
S is a finite subset of X containing the singular set of X;
U = X - s.
License or copyright restrictions may apply to redistribution; see https://www.ams.org/journal-terms-of-use
INDEX THEORY ON CURVES 593
The local structures of X and <n are well known (see e.g. [12, p. 28]). The metric on
U is the restriction of the Fubini-Study metric from the ambient projective space.
For simplicity we frequently give X a metric that agrees with 77*(metric on U) except
possibly on a neighborhood of tr~l(S), where (metric on X) > 7r*(metric on U).
Frequently in what follows U will be identified with-n~l(U). It is in this sense that
functions, all of which are C-valued, or forms on X can be restricted to U. U and
Tt~l(U) have the same structure as complex manifolds. However, the metric, volume
form and »-operator on U may differ from the metric, volume form and »-operator
on tt~1(U), which come from those on X. Which metric, volume form and »-operator
are desired will be indicated by the space, U or 77-_1(l7) or X, that appears in
accompanying notation.
For V any open subset with smooth boundary of U or X, with the closure of V
taken in U or X respectively, we make the following definitions.
Definition 3.1. For/, g g Cx(V), (f, g)L2(V) = fvf A * g. _
Definition 3.2. For/, g g C°°(V), (/, g)„Hv) = \vf A * g + ¡vdf A *dg.
Denote by T¿ the complexified cotangent bundle and by T* the antiholomorphic
cotangent bundle.
Definition 3.3. For a, ß g C°°(V, T£), («, ß)L^KTi) = iV« A * ß-
Definition 3.4. For a, ß g Cx(V, f*), (a, ß)L2(y f., = jvcx A * ß.
Definition 3.5. L2(V) = completion of {/g C°°(V): (f,f)o{V) < oo} in the
norm given by (,)L2{y).
Definition 3.6. H\V) = completion of {/g C°°(K): (/, /)w>(|/| < oo} in the
norm given by (, )H>(y).
Definition 3.7. L2(V, T¿) = completion of {a g C°°(V, F¿): (a, a)L2(VTi) < oo}
in the norm given by (, )L2(VT.Y
Definition 3.8. L2(V, f*) = completion of {a g C°°(K, T*); (a, a)^^ < oo}
in the norm given by (, )Lity,f*)-
Clearly
Lemma 3.9. The spaces in Definitions 3.5-3.8 are separable Hilbert spaces with inner
products extended from those of Definitions 3.1-3.4, respectively.
Definition 3.10 Ho\l(V) = {functions holomorphic on V and contained in
H\V)}.
Lemma 3.11. One can define an equivalent norm on Hl(V) by starting with the inner
product given by fvf A * g + ¡vdf A * 9g + /K9/ A * 3g.
Let Y be an arbitrary open subset of X with smooth boundary and with
Y = w(Y) n U.
Lemma 3.12. For a g C°°(Y, T¿) and ß = restriction of a to Y, fyct A * ä = jYß
A *ß.
The proofs of Lemmas 3.11 and 3.12 are well known calculations.
Lemma 3.13. Restriction is an isometry L2(Y, f*) -* L2(Y, f*).
License or copyright restrictions may apply to redistribution; see https://www.ams.org/journal-terms-of-use
594 PETER HASKELL
Proof. Observe that C?(Y - w~\S) n Y,T*) and CCX(Y,T*) are dense in
L2(Y, T*) and L2(Y, T*), respectively, and apply Lemma 3.12.
Lemma 3.14. Restriction gives bounded linear maps L2(Y) -» L2(Y) and Hl(Y) -»
H\Y).
Proof. This lemma follows immediately from our choice of metrics, which in fact
makes these maps norm-decreasing, and Lemma 3.12.
Lemma 3.15. The metric on U is conical in the sense that each point in X has a
neighborhood in which each branch is quasi-isometric to a cone.
Proof. See [7, p. 323] or calculate directly.
Example 3.16. Most interesting aspects of this paper can be seen in the cuspidal
cubic x2z = y3, in CP2. Consider a coordinate chart centered at 7r_1(singular point).
Change to polar coordinates. For/supported near it "'(singular point),
(f,f)LHu)~f\ffr3drdd and (f,f)LhX)~j\f\2rdrde.
If |/ | behaves like 1/r near ^(singular point), /g L2(U) but /<£ L2(X) and
/<£ H\U).
4. The Laplace operator. In this section a selfadjoint Laplace operator is identified,
and the functional calculus is applied to it to prove Proposition 4.4, which is needed
in the succeeding sections. Many arguments in this section are true on U or X. In
statements that hold in either case, the space "Í/" or "X" is suppressed in the
notation. For example the Laplace operator on U is h.v and that on X is A^; and
when a statement is true of either operator, A will be used.
Cheeger [5, p. 93] shows that d with domain {/g C°° n L2: df g L2(T£)} is
closable and that 5 = -*d* with domain {w g Cx(T£) n L2(T¿): ôco g L2} is
closable. Note that the domain of d is H1.
Definition 4.1. A = - Sd as an operator L2 -» L2.
Since the metrics on U and X are conical, d* = 8 [7, p. 321], and by [13, p. 312].
Lemma 4.2. A is selfadjoint.
Thus we can apply to A the functional calculus arising from the spectral theorem.
By [13, p. 307] (1 - A)"1 is selfadjoint, has norm < 1 and has positive spectrum. We
will see shortly that range((l - A)"1) c Hl. By Corollary 5.10, which is independent
of arguments here, the inclusion Hl ■-» L2 is compact (see also [6]). Therefore, L2 has
an orthonormal basis of eigenvectors of (1 - A)"1 and thus of A. Also the spectrum
of A is a discrete infinite set of nonpositive real numbers, each an eigenvalue of finite
multiplicity. This information allows one to justify each application of the functional
calculus that follows.
Let J be the inclusion H1 •-* L2. Since J is bounded as a map H1 -» L2, its adjoint
J* is defined everywhere. Thus for all/ g L2 and g g H1,
(f,Jg)r- = (J*f,g)„' = (J*f,g)L2+(dJ*f,dg)L2(Ti)=((l-\)J*f,g)L2.
License or copyright restrictions may apply to redistribution; see https://www.ams.org/journal-terms-of-use
INDEX THEORY ON CURVES 595
Since Jg = g, J* = (1 - A)"1. It follows that J's partial isometry part, which is an
isometry, is (1 - A)1/2.
Lemma 4.3. (1 - A)1/2: H1 -* L2 is an isometry. I.e. Hl = domain((l - A)1/2).
Let W be the subset of X for which the metric on X agrees with 77-*(metric on U).
In what follows the support of an L2 or Hl function refers to the intersection of
supports of functions in the equivalence class.
Proposition 4.4. For arbitrary e > 0, there exist maps Le: L2(U) -» Hl(U) and
Le: L2(X) -* H\X) satisfying:
(a) for u G support(Le(/)), distance^, support(/)) < e; for x G support(L£(g)),
distance(x, support(g)) < e;
(b) (1 - Ay)"1/2 - Lc: L2(U) -» Hl(U) is compact; (1 - Aky1/2 - Le: L2(X)
-* Hl(X) is compact (in fact these operators are continuous between domains of any
two powers of 1 — Auorl — A % respectively);
(c) iff G L2( X) has support contained in [x g X: distance(x, X — W) > e}, then
LE(f) = Le(f) (here we identify functions by the restriction convention o/§3);
(d) Le: L2(U)^ H\U) and Le: L2(X) -* Hl(X) are isomorphisms of topological
vector spaces.
The rest of this section is a proof of this proposition. Parts (a), (b) and (c) are
taken from [8].
Let X be the spectral representation of (1 - A)1/2, regarded as an unbounded
operator from L2 to L2. By change of variable and [10, 858.801],
(4.5) ^ = ff%.yV •'o y '
Let 4>e(y) be a monotonically decreasing C00 bump function on [0, oo] that is
identically 1 in a neighborhood of 0 and that has support contained in {y:
0 < y < e/2}. Then
(4.6) f f ^Idy = f /" Z&p-M,) dyV77 •'o yw VW •'o y '
\¡2 rx cos(Xy) ,„ , s\ ,+ ̂ f \/2 (l~^(y))dy.
V7T •'o y '
Lemma 4.7. For any ieR there is an operator Tk, bounded on the domain of any
power of(l - A), such that (\/2 / \¡Tr)f¿°(cos(\y)/yl/2) (1 - <í>E(vO) dy is the spectral
representation of(l — A)k °Tk.
Proof. Integrate by parts.
To understand the properties of the operator represented by
i/2~ /-°° cos(M , ,~rJ —Í7Í— <t>Ày)dy,
consider the hyperbolic equation (32/3/2 + 1 — A)/(z, /) = 0. For initial condi-
tions f0(z) = f(z,0) g L2(z-space) and (d/dt)f(z, t)\l=0 = 0, the solution of this
License or copyright restrictions may apply to redistribution; see https://www.ams.org/journal-terms-of-use
596 PETER HASKELL
equation is cos((l - A)1/2?)(/0(z)) [14, p. 70ff.]. When the domains of powers of the
Laplacian are locally defined, this solution exhibits finite propagation speed [14, p.
79ff.]:
(4.8) forz g supportícos((l - A)1/2í)/0(z)J, distance(z,support(/0)) <|?|.
By (4.5) and (4.6)
t. r.\ -v i \f2 rx cos(Av) , v , \¡2 r00 cos(\y) . . ., ,(4.9) X~l = "—\ \*,<?e(y)dy + — 1/2 v1 - *Áy)) dy •
yV Jo y ' V77 Jo y '
Let Lf be the operator represented by [(^2 / ]fJr)j™(cos(Xy)/y1/2)<pi!(y) dy]2 where
X is the spectral representation of (1 - Av)1/2. Let Le be the operator represented by
[(y/2 / \^r)J™(cos(\y)/y1/2)(i>t(y) dy]2, where X is the spectral representation of
(1 — A^)1/2. Since the domains of powers of these Laplacians are locally defined,
the condition on support^) and (4.8) imply Proposition 4.4(a). Lemma 4.7 implies
Proposition 4.4(b).
Finite propagation speed implies uniqueness of solutions [14]. In particular for/as
in Proposition 4.4(c), cos((l - A[/)1/2/)(/) = cos((l - A^)1/2i)(/) for 0 < t < e
because each can be regarded as the unique solution to the hyperbolic equation on X
for 0 < / < e. Proposition 4.4(c) follows.
By Lemma 4.3 and change of variable, s = Xy, to prove Proposition 4.4(d) it
suffices to show that there exist finite positive kx and k2 such that for each value of
X, also denoted X,
cos(s)/s1/2 ds/ \ (cos(î)A1/2)<î>f(.î/à) ds < k2.0 JQ
By [10, 858.801] (y/2 / )fn:)Jxfcos(s)/sl/2 ds = 1. Let ,4 = sup{.y: <¡>e(y) = 1} and letB = sup{>>: <j>c(y) 4= 0}. Let F(r) = j¿cos(s)/sl/2 ds. Let Xx = smallest value of X
such that (w/2)/X < B. Then for X < Xx
(4.11) F(A)^f (cos(s)/sl/2)<P[(s/X)ds < F(tt/2)■'o
and for X > Xx
Jr>00(cos(s)/s1/2)<pe(s/X) ds < F(ir/2).
o
One checks (4.11) and (4.12) by using the alternating sum associated with each
integral by virtue of the behavior of cosine. By [15, p. 55, 745] (4.11) and (4.12)
imply (4.10).
Corollary 4.13. Statements analogous to those in Proposition 4.4 are true for
(1 - A)1/2. (For (d) the maps are H[ -* L2.)
Proof. (1 - A)1/2 = (1 - A)°(l - A)"1/2 = (1 - A)°(L(;) + ((1 - A)"1/2 -
L(;')) and 1 - A is local.
License or copyright restrictions may apply to redistribution; see https://www.ams.org/journal-terms-of-use
INDEX THEORY ON CURVES 597
5. Defining a cycle.
Theorem 5.1. Let X be a complex projective algebraic curve. Let U = X — S, where
S is a finite subset of X which contains the singularities of X. Let\p0: C( X) —> ¿zf(L2(U))
be multiplication of L2 functions by restrictions of continuous functions. Let \px:
C(X) —> =S?(L2(t/, T*)) be multiplication of L2 forms by restrictions of continuous
functions. Let 3 be the natural extension of the ^-operator from Cx H1 functions to Hl
functions. Then (L2(U), \p0, L2(U, T*), ^,3 °(1 — A,y)~1/2) is a cycle representing
an element of Kq(X).
Proof. By Lemma 3.9 the proof reduces to proofs of three propositions.
Proposition 5.2. Give C(X) a C*-algebra structure with conjugation as * and the
sup norm as norm. Then \p0 and\px are unital algebra *-homomorphisms.
Proof. Since X is compact, each element of C(X) has finite sup, and the
proposition follows immediately from definitions.
Proposition 5.3. 9°(1 - A^)"1/2: L2(U) -+ L2(U, f*) is a bounded Fredholm
operator.
Proof. By Lemma 4.3 it suffices to show that 3: Hl(U) -* L2(U, T*) is bounded
and Fredholm. Lemma 3.11 shows that 9 is bounded. Recall that w: X -» X is the
desingularization of X. By compactness of X, the 9-Poincaré lemma for currents [11,
p. 385] and elliptic regularity, X can be covered with a finite collection of open sets
{ W¡} such that for each i:
(a) dW¡r\S= 0;
(b) 0 -> HolVH^i)) -* H\v-\W,)) ^ L2(-n-\W,), f*) -> 0 is exact.
Denote by W an arbitrary one of the W¡. Consider the commutative diagram
(5.4)
0 -> Uoll(tr-1(W)) - Hl(ir-\W)) ^ L2(tt\W),T*) -* 0
4'"1 4/ *^2 4-^3
0 -^ no\\wr\u) -» Hl(wc\u) -» L2(wn u,T*) -» o
where the vertical maps are restrictions (under convention of §3) of functions or
forms. By Lemmas 3.13 and 3.14 the vertical maps are continuous and R3 is an
isometry.
Lemma 5.5. The bottom row of(5A) is exact.
Proof. Elliptic regularity determines the kernel of 3. A diagram chase shows that
3 is surjective.
Lemma 5.6. Pj is a bijection.
Proof. If K is a compact subset of X that is disjoint from S, the norms on Hl(X)
and Hl(U) are equivalent on functions with support in K. Thus this lemma reduces
License or copyright restrictions may apply to redistribution; see https://www.ams.org/journal-terms-of-use
598 peter haskell
to the statement: if / has a pole or essential singularity at a point in tt"1(5'),
/ <£ Hl(U). This statement can be checked by using the norm described in Lemma
3.11 with special attention paid to the term involving 9/.
The five lemma implies that P2 is bijective. The open mapping theorem implies
that P 2 is an isomorphism of topological vector spaces. One can then use a partition
of unity subordinate to the ir~l(W¡) to prove
Lemma 5.7. Restriction gives an isomorphism of topological vector spaces Hl(X) -»
H\U).
Consider the commutative diagram
H\X) -> L2(X,f*)
(5.8) 1 1
Hl(U) ^ L2(U,f*)
where each vertical map is restriction. By the 9-Poincaré lemma for currents and
elliptic regularity the top map is Fredholm and has index equal to the Euler
characteristic of H$'(X). It follows by Lemmas 3.13 and 5.7 that 9: Hl(U) -*
L2(U, T*) is Fredholm, which finishes the proof of Proposition 5.3, and that
Corollary 5.9. The index ofd: Hl(U) -» L2(U, T*) equals the Euler characteristic
ofHf(X).
The arguments above help establish the following (see also [6]):
Corollary 5.10. The inclusion H\U) ■-» L2(U) is compact.
Proof. Consider the commutative diagram
Hl(X) ¿ L2(X)
(5.11) T*. lh2
Hl(U) -£ L2(U)
where ix and <2 are inclusions, hx is (restriction)"1, and h2 is restriction. Apply
Rellich's lemma on X.
Proposition 5.12. For each f g C(X), 3 °(1 - Ay)"1/2 ° t//0(/) - ^(/)»3 °
(1 - A^)"1/2: L2(U) -> L2(U, f*) is compact.
Proof. For simplicity let D denote 3°(1 - A^)"1/2: L2(U) -» L2(U, f*). The
finite set n~l(S) is denoted {sA eJ. All distances come from the metric on U.
The map C(X) "-* C(X) given by f '-* f ° ft leads to extensions of \p0 and \px to
C(X), both of which are unital algebra *-homomorphisms. By abuse of notation,
both extensions will be denoted by m. That is, / g C(X) goes to mf g ¿£(L2(U))
and w/gjS?(L2(l7, F*)).
Choose an arbitrary/ g C(X). Choose {gn }„ez,n>o sucn mat for each n:
(a)?,eq!);
License or copyright restrictions may apply to redistribution; see https://www.ams.org/journal-terms-of-use
INDEX THEORY ON CURVES 599
(b)sup|g„ - f°ir\ < 1/n.
(c) for each/ g J there exists 5 . > 0 such that gn is constant in a neighborhood of
s, with radius at least 8nj.
A standard construction allows such choices. Condition (b) implies that mgn -» w/o„
in operator norm. Since D is bounded, compactness of D ° mg -* mg ° D for each n
implies compactness of D °\$/0(f) - iix(f)° D. Choose an arbitrary element of {g„}
and call it g. Call its 5 -'s S-'s.
Lemma 5.13. D ° mg - mg° D = 3 °(1 - Ay)"1/2°mg - 3°wg°(l - Ay)_1/2 +
9 o mg »(I - Ay)"1/2 - mg o 3 o(i - Ay)"1/2.
Proof. A middle term has been subtracted and added. Because g G CX(X) and
dg = 0 in a neighborhood of 7r_1(S), mg: Hl(U) -* Hl(U) is a bounded linear map
and the middle term is well defined.
Lemma 5.14. 3 ° mg°(l - Ay)"1/2 - mg°3 °(1 - Ay)"1/2: L2(U) -» L2(U,f*)
is compact.
Proof. By Lemma 4.3 this lemma reduces to: 3°wg-wg°3: Hl(U) -*
L2(U, T*) is compact. That 9 ° mg - mg ° 3 = m¿g is seen by applying the product
rule to (3 ° mg — mg ° 9) restricted to C°°(U) fï HX(U) and extending by continuity.
By choice of g, 9 g g C™(U, T*). Therefore, denoting by ( , )u pointwise evaluation
of the metric on Fc*, we have supueu(dg(u), dg(u))1/2 < oo, and for all h G L2(U)
(5.15) (dg • A.og • M^.f., « f sup (ogíiO.Sgí«))1/2) -(A, A)1^.
Therefore, the domain w5g can be extended to L2(U) and so m%g: Hl(U) -*
L2(U, T*) is compact.
Lemma 5.16. 3 °(1 - Ay)"1/2 ° mg - 3 ° mg °(1 - Ay)"1/2: L2([/) -* L2(£/, f*)
ii compact.
Proof. Since 3: /F(c7) -^ L2(U, f*) is bounded, it suffices to show that
(1 - AuYl/2°mg- mg°(l - Ay)"1/2: L2(U) -+ H\U) is compact. We prove the
equivalent statement:
(5.17) wg-(l - Ay)1/2oWgo(l - Ay)"1/2:L2(i/)-> L2({/) is compact.
For convenience let L denote 1 - Ay.
Choose e > 0 such that 5e < min .^(5,.). Choose a C°° partition of unity /¿0 and
¡ij for all/ G J satisfying:
(a) distanceísupportífto), tt~1(S)) > 3e;
(b) distance^ — support(/i •), s¡) < 4e for each/ g J.
(Remember distances come from metric on U.)
(5.18) mg- Ll/2°mg°L-l/2= (mg- L1'2 ° mg° L~l/2)° m^
+ £ (mg- Ü'2 ° m g° L~x/2)° m¡l .
jej
License or copyright restrictions may apply to redistribution; see https://www.ams.org/journal-terms-of-use
600 peter haskell
Let Kj be the constant fonction with value equal to g's constant value in the 8-
neighborhood of s¡. Then for each/ G J
(5.19) (mg - F1/2 » mg ° L"1/2) ° m^ = (m,.^ - L1/2 ° «rIj ° L"1/2) <> mPj
+ {mKi-Ll/2omKoL-'/2ym^.
Linearity of powers of L implies that the last term of (5.19) is zero. By condition (b)
on the partition of unity, mg_K ° m^ = 0. Therefore
(5.20) (wg- Ü/2°mg°L-x/2)°m^¡ = -Ll/2 ° mg_K ° L~l/2 ° m .
Replace L~l/2 by Le + (L~l/1 - LE) (see Proposition 4.4). By condition (b) on the
partition of unity and by Proposition 4.4(a), m K ° L^m^ =0. By Proposition
4.4(b) Ll/2°mg_K °(L~1/2 - Le)°m maps L2(U) continuously to Hl(U). By
Corollary 5.10 it is compact as a map to L2(U).
To prove Lemma 5.16 only compactness of (m — L1/2 ° mg ° L~l/2)° m^ re-
mains to be shown. Choose another partition of unity v0 and v¡ for all je7
satisfying:
(a) each v¡ is constant on components of a neighborhood of <n~l(S);
(b) distance(support(i>0), tt~1(S)) > 2e;
(c) distance(support(ii0), support(fy)) > e for each y g /.
(5.21) (m g - Lx/2°mg°L~l/2)°m^ = mro° mg° m^ + £ mv, ° mg° m^0
— L1/2°mv ° m„° L~1/2 ° m„VQ £ Mo
— >Z L1/2 ° m„ » m ° L~l/2 ° m„ .*-^ vj S Mo
yey
Condition (c) on the y-partition of unity implies that mv ° mg° m = 0 and
Ll/2 ° m„ ° mg° L~ï/2 ° m = Ll/2 ° m„ ° mg°(L~l/2 - Le)° m^ for each/. Condi-
tion (a) on the ^-partition of unity, Proposition 4.4(b) and Corollary 5.10 imply that
this last expression is compact.
Let h = t>0 ■ g. It remains to be shown that mh ° m — L1/2 ° mh° L~1/2 ° m^:
L2(U) -* L2(U) is compact. Let N = {u g U: distance^, S) > e}. Extend the
metric on N to one on X. Call the nearly local operators of Corollary 4.13 L'E and L'e.
As maps from L2( X) to L2( X)
(5.22) m„ o mMo - (1 - A*)1/2 . mh .(1 - A*)"1/2 o „^
-»»•«^-(O - A^-)1/2 - ¿'Jo m^ °(1 -Aj)"'72.^
-i>i"((l - A^)"1/2- Le)oOT)io- L'eomhoLtomiio.
The left side of (5.22) is compact by the theory of pseudodifferential operators on
closed manifolds and Rellich's lemma. The second and third terms of the right side
of (5.22) are compact by Proposition 4.4(b) and Corollary 4.13. Thus
(5.23) mh°mv0 - L'c° mh° Lf° m^. L2(X) -> L2(X) is compact.
License or copyright restrictions may apply to redistribution; see https://www.ams.org/journal-terms-of-use
INDEX THEORY ON CURVES 601
The operator of (5.23) actually maps L2(N) to L2(N). Rewrite (5.22) with X
replaced by U, L'E by LE, and LE by Le. Once again the second and third terms on the
right are compact by Proposition 4.4(b) and Corollary 4.13. By Proposition 4.4(c)
and Corollary 4.13 the rewritten version of the first and fourth terms equals the
original version as a map of L2(N) to L2(N). Thus by (5.23) it is compact, and the
left side of rewritten (5.22) is compact.
6. A corresponding topological cycle.
Theorem 6.1. Let conditions and notation be as in Theorem 5.1. The element of
K¿(X) associated to (L2(U),t0, L2(U,T*),^x,d °(1 - Ay)"1/2) by the isomorphism
K¿( X) -* Kq(X) is represented by (X, 1, it), where it: X -» X is the desingularization
of X and 1 is the trivial line bundle X X C on X.
Proof. Choose neighborhoods of S, Ax and A2, with Ax c interior(/l2) and with
the intersection of X — A 2 with each irreducible component of X nonempty. Choose
e > 0 so that distance^, X - A2) > 4e. (Distance comes from metric on Í7.)
Choose an open set B such that:
(a)Ax c X - B czA2;
(b) distance^!, B) > e;
(c) distance^ - B, X - A2) > 3e.
Choose a set A3 such that:
(a)I-ici,c A2;
(b) distance^ - B, X - A3) > e;
(c) distance(^43, X — A2) > 2e.
Choose a set A4 such that:
(a) A3 c A4czA2;
(b) distance(^43, X — A4) > e;
(c) distance(v44, X — A2) > e.
Choose a set A5 such that:
(a) A4 cz A5cz A2;
(b) distance(^4, X - A5) > e;
(c)A5 c interior(y42).
Give X a metric that agrees with 77*(metric on U) except possibly on 7r~1(^41). Let
{^i}/e/ De tne set °f irreducible components of X. Let tt¡: X¡ -» X¡ be the desingu-
larization of each X¡. Let {Xk}keK be the set of connected components of X. For
each k let <xk be a fixed map of a point into Xk. Again let 1 denote the trivial line
bundle.
Lemma 6.2. As an abelian group K'0(X) is freely generated by {(X¡,1, w,)}lS/ U
{(point,l,ak)}keK.
Proof. Use Mayer-Vietoris sequences and compare to H*(X).
Lemma 6.3. An element ß of K'0(X) is determined by {p*([E] n ß): [E] is an
element of K°(X) represented by a bundle E of fiber dimension one or zero on X}.
Herep* is the map K0(X) -» K0( point) = Z induced by p: X -* point.
License or copyright restrictions may apply to redistribution; see https://www.ams.org/journal-terms-of-use
602 PETER HASKELL
Proof. For ß = \Lk^Kak(point, l,ak) + £/e/6;(À'„l, irt)] and F a vector bundle
on X with dk(E) = fiber dimension of E over Xk,
p*([E]nß)= L"k-dk(E)+ 2Zbi(ch(TT,*E)uTd(TX,))[Xi]keK iel
(see §2). For each k g K let Ek = Xk X C. For each pair (z, k) g / x K such that
X¡ c X4, define Eik to be the bundle satisfying:
(a) fiber dimension of Eik is one over X- and is zero over each Xk with k ¥= k;
(b)c1«F^)=l;
(c) cx(tt*E^ = 0 for i * i.
Using only {Ek}keK and {Eik} as defined above, one can form enough linear
equations to determine all of the ak and b¡.
Note that for F a vector bundle on X that occurs as the image of a family of
projection matrices acting on C" (see §2), -n*E occurs as the image of the pullback
under it of the family of projection matrices. Note also that each element of K°(X)
described in Lemma 6.3 can be associated with a family of projection matrices
that at every point in A 2 have 1 or 0 as upper left entry and zeros elsewhere. Call
such a family distinguished. Recall from §2 that (X,l,m) is associated to
(L2(X),$0,L2(X,T*),^x,d°(l - A *)"1/2), where ¿,(/) = multiplication by / ° it.
Switching to the cap product and push-forward in the analytic theory (see §2), we
see by Lemma 6.3 that Theorem 6.1 is implied by the following: for each [F] of
Lemma 6.3, with P0, Px, P0, Px formed from a distinguished family of matrices for F
and \p0, \px, t//0, \px respectively,
index[pio((3o(i - Ay)"1/2) ®/)°P0:
Po(L2(U) 9 C") - P1(F2(t/, f*) 9 C")]
= index[pi°((3°(l - Ak)~l/2) ®/)°P0:
P0(L2(X)® C") ^ px(l2(x, F*)® C")].
By Proposition 4.4(b) (6.4) is equivalent to
(6.5) index[P1°((3oFE) ® l)°P0] = index[Px °((3 » LE) 8/)»P0],
where the domains and ranges are as in (6.4).
Henceforth restrict attention to an arbitrary [F] and a distinguished family of
matrices for F, and denote by P and R the operators of (6.5). Ker(P) and Ker(P)
denote the kernels of R and P. Each kernel is finite dimensional. PB denotes
restriction to B. Restriction is a projection on a direct summand of any Hubert space
to which PB is applied. B is implicitly identified with ir~1(B). Note that as long as we
work with Hubert space elements supported in A4 or in tr~1(A4), R can be viewed as
3°LE: L2(U)^ L2(U,T*) and R asd ° Le: L2(X) ^> L2(X,T*).
Lemma 6.6. dim(Ker(F)) = dim(Pß(Ker(P))).
Proof. Pb: Ker(P) -» PÄ(Ker(P)) is surjective by definition.
License or copyright restrictions may apply to redistribution; see https://www.ams.org/journal-terms-of-use
INDEX THEORY ON CURVES 603
Assume/, g g Ker(P) and PB(f) = PB(s)- Then support(/ - g) cz U - B (sup-
port is defined in the paragraph preceding Proposition 4.4).
0 = R(f) - R(g) = R(f-g) = (9°L£)(/- g) = 9(LE(/- g)).
Since support(FE(/ - g)) c A3, Le(f - g) is a holomorphic function in Hl(U) with
support in interior(^42 n U). Therefore, Le(f - g) = 0, and so/- g = 0. Thus PB:
Ker(P) -> PB(Ker(P)) is injective.
Analogous arguments on X give
Lemma 6.7. dim(Ker(P)) = dim(PB(Ker(P))).
Lemma 6.8. ///g PB(Ker(P)), there exists a unique g g L2(7r_1(^4)) c L2(X)
such thatf + g G Ker(P).
Proof. Existence: /g PB(Ker(P)) => there exists h g L2(í/ - B) such that 0 =
R(f+ h) = P(/) + R(h). R(h) = -P(/). support(LE(A)) c A3. Therefore, there
exists g g L2(7r~l(A4)) with Le(g) = Le(h). R(g) = R(h) = -R(f). Proposition
4.4(c) implies that R(f)=R(f). Thus 0 = R(g) + R(f) = R(g + /). g+/GKer(P).
Uniqueness: Suppose gx and g2 satisfy the above conditions on g. Then 0 = R(gx)
- R(g2) = P(gi — g2). Le(gx — g2) is holomorphic with compact support in
interiorÍTr-1^;,)). Le(gx - g2) = 0. gx - g2 = 0.
Lemma 6.9. dim(PB(Ker(P))) < dim(Ker(P)).
Proof. Lemma 6.8 defines a linear map from PB(Ker(P)) to Ker(P). We proceed
by showing that this map is injective.
Suppose we have fx and f2 in PB(Ker(P)) whose images under the above map,
/i + gx and/2 + g2, are equal. Thenfx = f2 on U - A4.
There exist hx, h2 g L2(U — B) such that fx + hx and /2 + h2 are in Ker(P).
Since
support(fx + hx-f2-h2)czA4 and R(fx + hx - f2 - h2) = 0,
Le(fx + hx - f2 — h2) is holomorphic with support in interior(/l2 DU). Thus it
equals 0 and so doesfx + hx — f2 — h2. It follows that/j — f2 = h2 — hx, and by the
nature of the supports of these functions,/! = /2.
Reversing the roles of P and P in the preceding arguments, one proves
Lemma 6.10. dim(PB(Ker(P))) < dim(Ker(P)).
Lemmas 6.6, 6.7, 6.9 and 6.10 give
Lemma 6.11. dim(Ker(P)) = dim(Ker(P)).
Lemma 6.12. Image(P) = Image(P).
(Equality refers to identification by restriction from X to U. Recall that the range
spaces of P and P are isomorphic as Hubert spaces under this restriction.)
License or copyright restrictions may apply to redistribution; see https://www.ams.org/journal-terms-of-use
604 PETER HASKELL
Proof. Let PV^B and P*_B denote restriction to U - B and X — B respectively.
domain(P) = PB(domain(P)) e Pu_B(domain(P)). domain(P) = PB(domain(P))
ffi Py_ B(domain(P)). PB(domain(P)) = PB(domain(P)) and P restricted to
PB(domain(P)) equals P restricted to PB(domain(P)). w g R(PL/_B(domain(R)))
=> there exists wx g Hl(U) with support in A3 O U such that dwx = w => there
exists w2 g L2(tt~1(A4)) such that Le(w2) = wx => w g Image(P). An analogous
chain of implications shows that P(Py_B(domain(P))) c Image(P).
References
1. M. Atiyah, Global theory of elliptic operators, Proc. Internat. Conf. on Functional Analysis and
Related Topics, Univ. of Tokyo Press, Tokyo, 1970, pp. 21-30.
2. _, K-theory, Benjamin, New York, 1967.
3. M. Atiyah and I. Singer, The index of elliptic operators. I, Ann. of Math. 87 (1968), 484-530.4. P. Baum and R. Douglas, K homology and index theory, Proc. Sympos. Pure Math., vol. 38, part 1,
Amer. Math. Soc., Providence, R. I., 1982, pp. 117-173.
5. J. Cheeger, On the Hodge theory of Riemannian pseudomanifolds, Proc. Sympos. Pure Math., vol. 36,
Amer. Math. Soc., Providence, R. I., 1980, pp. 91-146.
6. _, On the spectral geometry of spaces with cone-like singularities, Proc. Nat. Acad. Sei. U.S.A.
76(1979), 2103-2106.
7. J. Cheeger, M. Goresky and R. MacPherson, L2-cohomology and intersection homology of singular
algebraic varieties, Seminar on Differential Geometry (S.-T. Yau, ed.), Princeton Univ. Press, Princeton,
N. J., 1982, pp. 303-340.
8. J. Cheeger and M. Taylor, On the diffraction of waves by conical singularities. I, Comm. Pure Appl.
Math. 35(1982), 275-331.
9. R. Douglas, C*-algebra extensions and K-homology, Princeton Univ. Press, Princeton, N. J., 1980.
10. H. Dwight, Tables of integrals and other mathematical data, Macmillan, New York, 1961.
11. P. Griffiths and J. Harris, Principles of algebraic geometry, Wiley, New York, 1978.
12. J. Milnor, Singular points of complex hypersurfaces, Princeton Univ. Press, Princeton, N. J., 1968.
13. F. Riesz and B. Sz.-Nagy, Functional analysis, Ungar, New York, 1978.
14. M. Taylor, Pseudodifferential operators, Princeton Univ. Press, Princeton, N. J., 1981.
15. G. Watson, A treatise on the theory of Bessel functions, Cambridge Univ. Press, Cambridge, 1966.
Department of Mathematics, Purdue University, West Lafayette, Indiana 47907
License or copyright restrictions may apply to redistribution; see https://www.ams.org/journal-terms-of-use